Evaluating Priority Dispatch for Premium Membership in On-demand Taxi-sharing

1 Introduction

• The study tests whether premium membership can operationally create a measurable time advantage, while leaving fare design open.The evaluation also asks when premium service substantially degrades non-premium service quality.Source: §1 · pp. 1–2
• The authors compare deadline shortening with a version that also reserves part of the fleet for premium users, using SUMO-based simulations.Both mechanisms prioritize premium users’ total travel time from demand generation to arrival.Source: §1 · p. 2
• Deadline shortening creates a travel-time gap mainly when the fleet is under pressure, while dedicated vehicles can create a clearer off-peak gap but reduce non-premium capacity under congestion.The appropriate mechanism and premium ratio depend on demand pattern and operating conditions.Source: §1 · p. 2

2 Related Work

• Ride-pooling research has quantified pooling potential and developed real-time optimization, but generally assumes homogeneous users rather than explicit service differentiation.Santi et al. quantified pooling potential using shareability networks, while Alonso-Mora et al. proposed a real-time optimization framework.Source: §2 · p. 2
• Prior work studies price–quality tradeoffs and differentiated service classes, but often focuses on pricing or theoretical design rather than real-time spatial dispatch.Examples include waiting-time differentiation, service quality contracts, and heterogeneous service classes.Source: §2 · p. 2
• This study evaluates dispatch-level priority before fare optimization, measuring both potential travel-time advantage and non-premium-user burden across demand patterns.The paper frames this as isolating the dispatch mechanism.Source: §2 · p. 2

3.1 Premium and Non-Premium Users

• Premium membership is evaluated by whether it reduces premium users’ total travel time while keeping the burden on non-premium users small.Premium users are modeled as time-sensitive travelers willing to accept an additional burden, such as a higher fare, for shorter travel time.Source: §3.1 · p. 3

3.2 Allocation Method

• SAVS dispatches through sequential optimal insertion, maintaining an ordered waypoint list for each vehicle without changing the order of existing waypoints.Figure 1 illustrates the method’s concept.Source: §3.2 · p. 3, Figure 1 · p. 3
• For each new demand, vehicles test feasible insertions of its origin and destination and calculate insertion cost as existing-waypoint delays plus the new demand’s expected completion time.The procedure is based on Noda et al..Source: §3.2 · p. 3
• Figure 2 presents the proposed dispatch methods, including recalculating premium requests’ allowable deadline interval before insertion.Source: Figure 2 · p. 4
• Insertions that cause an existing or new demand to miss its deadline are excluded, and the demand goes to the vehicle with the lowest minimum insertion cost.Source: §3.2 · p. 4

3.3 Efficiency Metric

• Time per distance (TPD) is the primary efficiency metric, calculated from waiting time and in-vehicle travel time divided by distance for completed trips.The analysis also measures the proportion of canceled demands among all generated demands.Source: §3.3 · p. 4
• The reported group metrics average each run’s mean TPD across 30 simulation runs; a desirable outcome lowers premium TPD without excessive growth in non-premium TPD.Source: §3.3 · p. 4

4 Proposed Dispatch Methods

• The baseline uses ordinary sequential optimal insertion, retains premium labels only for evaluation, and applies the same deadline to both user groups.Source: §4 · p. 4

4.1 Shorter-Deadline

• Shorter-Deadline prioritizes premium requests by tightening their allowable time constraint through an increased slack parameter vprem.The premium deadline uses a lower-bound adjustment involving average vehicle speed and N; increasing vprem strengthens priority, while excessively strict deadlines may increase cancellations.Source: §4.1 · p. 5

4.2 Dedicated-Fleet

• Dedicated-Fleet extends Shorter-Deadline by reserving vehicles for premium users; premium requests can use both vehicle groups, but non-premium requests can use only shared vehicles.This can make less-crowded vehicles available to premium users while reducing the vehicles available to non-premium users.Source: §4.2 · p. 5

5.1 Map, Fleet, and Simulation Horizon

• The experiments use a synthetic 15 × 15 grid with 400 m node spacing and a 1.6 km × 1.6 km central urban area.The full map covers 5.6 km × 5.6 km and is divided into entire-area and urban-area zones.Source: §5.1 · pp. 5–6, Figure 3 · p. 6
• Baseline and Shorter-Deadline share all 12 vehicles; Dedicated-Fleet reserves a ratio-based portion for premium users, with premium labels assigned randomly.Each condition uses 30 random seeds, and methods are compared under matching conditions and seeds.Source: §5.1 · p. 6, Table 1 · p. 6

5.2 Demand Scenarios

• S1 has uniform demand across the area, S2 adds a peak-period increase, and S3 concentrates demand centrally with additional urban peak demand.Source: §5.2 · p. 6, Table 2 · p. 7

5.3 Evaluation Metrics

• Evaluation compares premium-user improvement and non-premium-user degradation in TPD against baseline, and reports the service gap δ between the groups’ mean travel times.Positive values indicate lower premium-user TPD and higher non-premium-user TPD.Source: §5.3 · p. 7

6.1 Effect of Shorter-Deadline and Dedicated-Fleet

• Shorter-Deadline creates a premium/non-premium TPD gap during peak periods, but δ becomes small off-peak when vehicle capacity is sufficient relative to demand.Source: §6.1 · p. 7, Figure 4 · p. 8
• In the steady scenario, Dedicated-Fleet increases δ off-peak by restricting non-premium users to shared vehicles while premium users can access the entire fleet.Source: §6.1 · pp. 7–8, Figure 5 · p. 8

6.2 Average Time per Distance for the Two User Groups

• In time-varying and urban-focused scenarios, higher premium ratios reduce remaining premium benefit and increase non-premium burden, especially with Dedicated-Fleet.At vprem = 0.5, the figure reports 30-seed means and 95% confidence intervals for both groups under each scenario and method.Source: §6.2 · p. 8, Figure 6 · p. 9

6.3 Cancellation Rate and Improvement-Rate Space

• A 5% limit on non-premium TPD increase permits approximately 20% premium time-per-distance reduction; a 15% limit permits approximately 30% among tested points.The trade-off varies by scenario, method, and premium ratio; the figure’s polylines connect different vprem values.Source: Figure 8 · p. 9, §6.3 · p. 10
• Shorter-Deadline cancellations tend to rise with vprem, while Dedicated-Fleet tends to keep cancellations lower; cancellation rates remain zero under all Steady conditions.Source: Figure 7 · p. 9, §6.3 · p. 10
• Strict Shorter-Deadline settings can produce a U-turn because cancellations make completed-trip TPD appear more favorable; Dedicated-Fleet avoids this pattern.The passage attributes the U-turn to excessive cancellations and TPD being computed only for completed trips.Source: §6.3 · p. 10

• Shorter-Deadline preserves non-premium access to the full fleet and is useful under pressure, but its effect is limited when capacity is sufficient.Source: §7 · p. 10
• The pattern appears across demand scenarios despite differences in spatial demand structure, supporting condition-dependent dispatch choices rather than one fixed rule.Source: §7 · p. 10
• Dedicated-Fleet can create clearer off-peak differentiation, but reduces effective non-premium capacity under congestion and is described as too rigid there.Source: §7 · p. 10

• Shorter-Deadline tightens premium deadlines in sequential optimal insertion, while Dedicated-Fleet also reserves part of the fleet for premium users.Source: §8 · p. 11
• Shorter-Deadline is useful under congestion, whereas Dedicated-Fleet is more suitable off-peak; the results do not establish universal benefit from premium membership.Source: §8 · p. 11
• Open issues include pricing, hybrid dispatch, non-random premium demand, additional demand scenarios, user-facing verification, and validation on other maps and fleet settings.Source: §8 · p. 11